Anthropogenic Drivers of Mangrove Loss and Associated Carbon Emissions in South Sumatra, Indonesia

: The Air Telang Protected Forest (ATPF) is one of the most dynamic and essential coastal forest landscapes in South Sumatra, Indonesia, because of its location between multiple river outlets, including the Musi catchment—Sumatra’s largest and most dense lowland catchment area. While most ATPF areas are covered by mangroves, these areas have been experiencing severe anthropogenic-driven degradation and conversion. This study aims to evaluate land cover changes and associated carbon emissions in the ATPF over a 35-year period (1985–2020) by utilizing the available Landsat and Sentinel imagery from 1985, 2000, and 2020. Throughout the analysis period, we observed 63% (from 10,886 to 4059 ha) primary and secondary forest loss due to land use change. We identiﬁed three primary anthropogenic activities driving these losses, namely, land clearing for plantations and agriculture (3693 ha), coconut plantations (3315 ha), aquaculture (245 ha). We estimated that the largest carbon emissions were caused by coconut plantation conversion, with total carbon emissions of approximately 14.14 Mt CO 2 -eq. These amounts were almost 4 and 21 times higher than emissions from land clearing and aquaculture, respectively, as substantial soil carbon loss occurs once mangroves get transformed into coconut plantations. While coconut plantation expansion on mangroves could generate signiﬁcant carbon stock losses and cleared forests become the primary candidate for restoration, our dataset could be useful for future land-based emission reduction policy intervention at a subnational level. Ultimately, our ﬁndings have direct implications for current national climate policies, through low carbon development strategies and emission reductions from the land use sector for 2030, as outlined in the Nationally Determined Contributions (NDCs).


Introduction
Mangrove forests are found in intertidal regions along tropical and subtropical coastlines [1] between 30 • N and 30 • S latitude [2,3]. These forests grow along muddy beaches with a low hydrodynamic intensity, especially in areas with expansive river estuaries and deltas where sediments are continuously supplied by upland catchments. The macroecological structure and species composition of mangrove forests are influenced by biogeographic, climate, and hydro-geomorphic factors [4]. Naturally, each mangrove species has its own level of tolerance to salinity concentration. This has implications for the zonation of a distinct species, from seaward to landward (microscale) factors, and hydro-geomorphic factors

Data Collection
The imagery data were obtained from Landsat 5 imagery for the years 1985 and 2000, and Sentinel-2 for the year 2020 (see Table 1 for the detailed satellite image information used in this study). Forestry thematic maps and relevant land cover maps obtained from the Provincial Forestry Service were used to support land cover classification and to improve the accuracy of the classification results. Additionally, a series of land cover ground checks were conducted in 2000 for primary mangrove forests, secondary mangrove forests, coconut plantations, fishponds, and open areas, as summarized in Table 2. The collected coordinates of each land cover were used for training points during the classification analysis and accuracy test. Coordinates were collected using a handheld Geographical Positioning System (GPS), as shown in Table 2. The pictures of each land use field condition obtained during the ground check fieldwork are presented and summarized in Figure 2.

Data Collection
The imagery data were obtained from Landsat 5 imagery for the years 1985 and 2000, and Sentinel-2 for the year 2020 (see Table 1 for the detailed satellite image information used in this study). Forestry thematic maps and relevant land cover maps obtained from the Provincial Forestry Service were used to support land cover classification and to improve the accuracy of the classification results. Additionally, a series of land cover ground checks were conducted in 2000 for primary mangrove forests, secondary mangrove forests, coconut plantations, fishponds, and open areas, as summarized in Table 2. The collected coordinates of each land cover were used for training points during the classification analysis and accuracy test. Coordinates were collected using a handheld Geographical Positioning System (GPS), as shown in Table 2. The pictures of each land use field condition obtained during the ground check fieldwork are presented and summarized in Figure 2.     For the carbon emissions estimate, we used secondary data of the mangrove carbon stocks and emission factors for the coconut plantation and aquaculture from previously published studies by [29][30][31][32]. Specifically, carbon stocks for the primary mangrove forest were obtained from a previous study performed in Sembilang National Park, approximatelỹ 25 km from the ATPF [29]. Here, protected primary mangroves store approximately 1319 ± 144 Mg C ha −1 of the total ecosystem carbon stocks [29]. It is composed of 320 ± 88, 20 ± 7, and 979 ± 152 Mg C ha −1 of total biomass, woody debris, and soil carbon stock pools, respectively [29]. While there are no specific carbon stocks data available for the secondary mangrove forest, we assumed their total biomass carbon stocks represent 76% of the primary forests [33]. Moreover, we used a carbon stock loss fraction of 88% for the total biomass and 95% for the soil carbon pools following mangrove conversion to coconut plantation, as previously documented [30,31] in the Philippines. Furthermore, carbon stock loss percentages of 83 and 52% for the total biomass and soil carbon losses, respectively, were used to calculate carbon stock loss due to aquaculture conversion, as reported in a global meta-analysis study [32].

Data Analysis
To link our analysis with an existing dataset, the official land cover category and definition were used and obtained from the national thematic map of the Ministry of Environment and Forestry of the Republic of Indonesia. The land cover type classification comprised primary forest, secondary forest, coconut plantation, open area, fishpond, and water body (Table 3 and Figure 2). We used the supervised classification method (visual interpretation) to perform image classification for all of the datasets. The visual interpretation method was an effective method for identifying land cover classifications [33]. While there was a spatial resolution difference between the Landsat and Sentinel products (Table 1), we used Landsat 8 image data acquired in 2020 to correct and adjust the land cover area generated by Sentinel-2 image data for the same year. This was done to improve the accuracy of the land cover classification results and to minimize the impact of the differences in the land use areas generated by all of the used datasets.
The land cover change detection analysis in ATPF was performed by overlaying the produced land cover maps between the periods studied. This technique is commonly used in remote sensing studies, and makes it possible to identify the extent of land cover changes between the given temporal period. The data analysis in this study was carried out using a Geographic Information System (GIS) with ArcGIS 10.5 software.
The estimation of the CO 2 emissions in this study focused on emissions generated by mangrove conversion to other land uses, namely coconut plantations, fishponds, and open areas (deforestation). Overall, the CO 2 emissions generated by each land use change between the analysis periods was estimated using the carbon stocks change approach [35]. First, we assumed that the initial carbon stocks of the primary mangrove cover in the first year of analysis were similar to the nearby mangrove forest reference in Sembilang National Park [29]. The initial carbon stocks in the secondary mangrove forests was 76% of the biomass carbon stock in the primary mangrove forests, which was previously reported in the national greenhouse gas inventory [33]. Subsequently, the carbon stock loss for each land use change was estimated by applying an emissions factor for the total biomass and soil carbon. The estimation of the CO 2 emissions is described in Equation (1).
where the CO 2 emission is the total CO 2 -eq emitted over the calculated land use change area, 44/12 is the conversion factor of carbon stocks to CO 2 -eq, carbon stock is the initial carbon stock for each land use, EF is the percentage of carbon stock loss generated by each land use, and LUC area is the area of land use change between periods obtained from the change detection analysis. We excluded an uncertainty calculation in this study because of data availability limitations. A summary of the CO 2 emission estimation for this study is provided in Supplementary Information Tables S1 and S2.

Mangrove Cover in 1985, 2000, and 2020
Overall, six major land cover types were identified that most recently dominated ATPF, including primary forest, secondary forest, coconut plantation, open area, fishpond, and water body ( Figure 3). According to the ground truth field data collection, the primary forest was dominated by true mangrove species, such as Nypa fruticans, Rhizophora apiculata, Sonneratia alba, Avicennia alba, Bruguiera cylindrica, Excoecaria agallocha, and Xylocarpus granatum, while the secondary forest was a mixture of mangrove stands and shrubs. carbon stock for each land use, EF is the percentage of carbon stock loss generated by each land use, and LUC area is the area of land use change between periods obtained from the change detection analysis. We excluded an uncertainty calculation in this study because of data availability limitations. A summary of the CO2 emission estimation for this study is provided in Supplementary Information Tables S1 and S2.

Mangrove Cover in 1985, 2000, and 2020
Overall, six major land cover types were identified that most recently dominated ATPF, including primary forest, secondary forest, coconut plantation, open area, fishpond, and water body ( Figure 3). According to the ground truth field data collection, the primary forest was dominated by true mangrove species, such as Nypa fruticans, Rhizophora apiculata, Sonneratia alba, Avicennia alba, Bruguiera cylindrica, Excoecaria agallocha, and Xylocarpus granatum, while the secondary forest was a mixture of mangrove stands and shrubs. We observed that the area of both primary and secondary forests decreased significantly over the period of 35 years (1985-2020), as summarized in Table 4. In 1985, the primary forest was 6257 ha (49.42% of total ATPF area), and after 35 years, the remaining primary mangrove forest area was only 2936 ha (23.19%). This change indicates a substantial loss of primary forest of more than 50% over the mentioned period. Similarly, the area of the secondary forest was also reduced significantly from 36.56% in 1985 to only 8.87% in 2020. In contrast, we identified a significant increase in open area (deforested forest), of nearly 50%, from the initial analysis year. Similarly, the coconut plantation area increased by 21% between 1985-2000, but decreased by 15% between 2000-2020. Finally, the fishpond area continued to increase during the entire analysis period of 1985-2020, despite the area of increase being only 189 ha or around 1.5% of the latest total ATPF area.  We observed that the area of both primary and secondary forests decreased significantly over the period of 35 years (1985-2020), as summarized in Table 4. In 1985, the primary forest was 6257 ha (49.42% of total ATPF area), and after 35 years, the remaining primary mangrove forest area was only 2936 ha (23.19%). This change indicates a substantial loss of primary forest of more than 50% over the mentioned period. Similarly, the area of the secondary forest was also reduced significantly from 36.56% in 1985 to only 8.87% in 2020. In contrast, we identified a significant increase in open area (deforested forest), of nearly 50%, from the initial analysis year. Similarly, the coconut plantation area increased by 21% between 1985-2000, but decreased by 15% between 2000-2020. Finally, the fishpond area continued to increase during the entire analysis period of 1985-2020, despite the area of increase being only 189 ha or around 1.5% of the latest total ATPF area. Note: ha-area of land cover type in ha; %-percentage area of land cover type of the total ATPF area.

Land Cover Change 1985-2000 and 2000-2020
During the first and second time periods, there was a decrease in the area of the primary forest by 935 and 2385 ha, respectively. During these same two time periods, secondary forests also experienced a decline of 2064 and 1440 ha, respectively ( Figure 4). Coconut plantations expanded by 2638 ha over the 1985-2000 period, but decreased by 1878 ha over the 2000-2020 period (Figure 4). Similarly, the deforested area expanded by 1878 ha in the second period. The fishpond area increased in the two periods compared with only less than 100 ha or 1% in 35 years. Similarly, a reduction in the water body area also occurred in these two periods (Figure 4).

Land Cover Change 1985-2000 and 2000-2020
During the first and second time periods, there was a decrease in the area of the primary forest by 935 and 2385 ha, respectively. During these same two time periods, secondary forests also experienced a decline of 2064 and 1440 ha, respectively ( Figure 4). Coconut plantations expanded by 2638 ha over the 1985-2000 period, but decreased by 1878 ha over the 2000-2020 period (Figure 4). Similarly, the deforested area expanded by 1878 ha in the second period. The fishpond area increased in the two periods compared with only less than 100 ha or 1% in 35 years. Similarly, a reduction in the water body area also occurred in these two periods ( Figure 4). The land cover change analysis indicated that in the 1985-2000 period, primary and secondary mangrove forests were mostly converted to coconut plantations ( Figure 5), with as much as 667 and 1970 ha, respectively. In the same timeframe, the degradation and conversion of primary forests into secondary forests and open areas were 361 and 178 ha, respectively. Additionally, only 66 ha of primary forest area was converted to fishponds.
In the period of 2000-2020, both primary and secondary forests experienced the largest degradation and conversion into open areas or deforestation ( Figure 5), by as much as 1844 and 1409 ha, respectively. Moreover, around 2556 ha of coconut plantations were also changed into open areas. In this second period, the conversion of primary forests to coconut plantations and fishponds were 290 and 110 ha, respectively. Meanwhile, the conversion of secondary forests into coconut plantations and fishponds were 386 and 40 ha, respectively.  The land cover change analysis indicated that in the 1985-2000 period, primary and secondary mangrove forests were mostly converted to coconut plantations ( Figure 5), with as much as 667 and 1970 ha, respectively. In the same timeframe, the degradation and conversion of primary forests into secondary forests and open areas were 361 and 178 ha, respectively. Additionally, only 66 ha of primary forest area was converted to fishponds.

Carbon Emissions Generated by Mangrove Conversion to Other Land Uses
Throughout 1985-2020, we observed that the largest mangrove loss in ATPF was driven by deforestation (3693 ha), followed by coconut plantations (3315 ha) and fishponds (245 ha; Figure 6 and Table 5). Coconut plantations were the primary mangrove loss driver before 2000, while deforestation largely occurred after 2000 (Figure 6a). However, we estimated the largest carbon emissions were from coconut plantation conversion throughout all of the analysis periods, with the total carbon emissions being approximately 14.14 Mt CO2-eq (Figure 6b). These amounts were almost four times higher than the emissions from deforestation, even though they impacted a larger area of mangrove loss ( Figure 6). During 1985-2000 alone, mangrove conversion to coconut plantations generated the largest carbon emissions, around 11.23 Mt CO2-eq or 94% of the total emissions over this period. Moreover, the largest carbon emissions from mangrove deforestation were 3.42 Mt CO2-eq, occurring during 2000-2020. We observed a low area change impacted by fishpond development in ATPF throughout the analysis years.

Carbon Emissions Generated by Mangrove Conversion to Other Land Uses
Throughout 1985-2020, we observed that the largest mangrove loss in ATPF was driven by deforestation (3693 ha), followed by coconut plantations (3315 ha) and fishponds (245 ha; Figure 6 and Table 5). Coconut plantations were the primary mangrove loss driver before 2000, while deforestation largely occurred after 2000 (Figure 6a). However, we estimated the largest carbon emissions were from coconut plantation conversion throughout all of the analysis periods, with the total carbon emissions being approximately 14.14 Mt CO 2 -eq (Figure 6b). These amounts were almost four times higher than the emissions from deforestation, even though they impacted a larger area of mangrove loss ( Figure 6). During 1985-2000 alone, mangrove conversion to coconut plantations generated the largest carbon emissions, around 11.23 Mt CO 2 -eq or 94% of the total emissions over this period. Moreover, the largest carbon emissions from mangrove deforestation were 3.42 Mt CO 2 -eq, occurring during 2000-2020. We observed a low area change impacted by fishpond development in ATPF throughout the analysis years.

Coconut Plantation as the Main Driver of Mangrove Loss in South Sumatra
The degradation of mangrove forests in the ATPF over the past 35 years was driven by various anthropogenic activities, including coconut plantations, fishponds, settlements, and agriculture. Our findings are consistent with regional-and global-scale assessments, according to which anthropogenic activities such as land-use change ares the primary driver of mangrove loss [15,17,25,[36][37][38][39][40][41][42][43][44][45][46]. Unlike other studies in the region that found aquaculture and oil palm plantations to be the dominant drivers of mangrove loss [36,46], we observed that coconut plantations were the second dominant driver in ATPF, after deforestation. This trend was also observed elsewhere in the Southeast Asia region, where coconut plantations also replaced some mangrove areas in the Philippines, with a substantial impact on reduced carbon stock [30,31]. According to the information from local communities, the coconut plantations ATPF began in 1972, which was promoted by the Bugis (Sulawesi) people who migrated to this area. Subsequently, most of the new coconut plantation areas were developed by replacing primary and secondary forests. Our findings suggest that coconut plantations are the new frontier of mangrove loss drivers, in addition to oil palm plantations, which were previously observed in the northern part of Sumatra [46].
We observed a large area of deforested mangroves or open area land cover type. According to the information from local communities obtained during fieldwork, this open area represents cleared lands owned by coastal communities and companies for various purposes, including the development of agricultural land and oil palm plantations.
Additionally, this open area was also used to regrow coconut trees, following a rejuvenation program after more than 20 years of the first plantation. This is consistent with our findings of ATPF's land cover in 2000, according to which coconut plantations occupied approximately 50% of the total area of ATPF. Furthermore, we observed a low number of fishpond expansion in ATPF, despite the typical land use commonly observed replacing mangroves in Southeast Asia [46].
The area of both primary and secondary forests in ATPF continued to decline during the 1985-2020 period. Nevertheless, the remaining forest areas were dominated by Nypa fruticans, spanning across almost all coastline areas of ATPF [22,47]. The domination of invasive mangrove palms such as Nypa fruticans indicates that the areas are heavily disturbed in ATPF. Researchers usually consider this species as one of the biological indicators for mangrove forest disturbance [47][48][49].

Carbon Emissions and Their Implication for Land Management and Restoration
The estimated carbon emissions suggest that coconut plantations generated four times more carbon than those from deforestation, despite deforestation being the largest driver of mangrove loss area in ATPF during the past 35 years. Our findings are consistent with previous studies in the Philippines, where soil carbon loss was dominant because of the significant coastal landscape transformation [11,30,31]. Planting coconut trees in the cleared mangrove area requires drainage in order to avoid high salinity water following tidal inundation. A similar modification has also also observed for the development of oil palm plantations on top of mangrove forests [46]. Therefore, such belowground modification and drainage may lead to elevated soil organic matter oxidation and dissolved organic matter release through aquatic pathways, resulting in lowered soil carbon stocks.
Despite the fact that we estimated the carbon emissions from secondary data sources, our estimates suggest that mangrove conversion to other land uses could generate larger emissions than the forests that only experienced deforestation. This is consistent with previous studies documented elsewhere [32,50], suggesting that avoiding future mangrove loss and conversion could lead significant emission reductions from the land sector. Our findings could be a proxy for land managers and policy makers on how to improve land management in coastal mangrove area, specifically by promoting conservation and restoration programs to support national emission reduction targets, as outlined in the Nationally Determined Contributions (NDCs). Some restoration programs in ATPF have been started since 2011 by the government and community of Banyuasin Regency. However, land tenure seems to be a persisting issue, as the converted land is owned by many entities, including local communities and private sectors. This situation has led to the restoration programs being placed on hold over the past decade. Additionally, the restoration of degraded mangrove forests might require a substantial budget allocation in the long term. Mukherjee [10] stated that habitat degradation as a result of anthropogenic activities and development, both in developed and developing countries, requires a restoration time of more than 20 years. In addition, the success of the restoration is highly dependent on the commitment of state institutions and on community support. Policy failure and the dysfunction of state institutions, as well as a lack of participatory awareness and community commitment, can disrupt the restoration and conservation process [51].
While the carbon emissions estimate in this study could indicate the magnitude and variation of emissions between land uses, the use of secondary carbon stocks data and stocks-difference approach may lead to several limitations. First, the secondary carbon stocks data obtained in 2010 from Sembilang National Park mangroves [29] may not be similar with the ones in ATPF, particularly when we these data were to estimate stocks back in 1985 and 2010 (e.g., space for time substitution approach). Second, mangrove carbon stocks and fluxes are highly varied toward spatial hydrogeomorphic setting and temporal variability, respectively [52]. Estimating carbon stocks and emissions should therefore carefully consider the spatial replication of field sampling. Future similar studies may also consider the use of shorter temporal satellite imagery acquisition, such as every 5 years, rather than 15 or 20 years. This is particularly important, because of the extensive dynamics of aboveground biomass in mangrove forests, with recovery rates up to 4-7 MgC ha −1 yr −1 [32]. Third, the singled-out emissions estimate using only the CO 2 equivalent neglects other greenhouse gases (GHGs), particularly (CH 4 ) methane [53], as well as aquatic carbon export. Therefore, our approach contributes further uncertainty regarding unaccounted emissions from CH 4 and aquatic components. While CH 4 is 28 times more potent compared with CO 2 [54], incorporating this GHG into future mangroverelated emission estimates will be essential in order to reduce uncertainty, specifically as it is reported that restoring and rewetting mangroves could generate higher CH 4 emissions, despite lowering CO 2 emissions [55].

Conclusions
We have identified that the land cover in the ATPF area is composed of primary forests, secondary forests, coconut plantations, open areas, fishponds, and water bodies. Our historical land use change reconstruction using remote sensing analysis further revealed that primary and secondary forest areas decreased continuously over a period of 35 years (1985-2020) in response to the aforementioned anthropogenic drivers. Specifically, we observed that the expansion of coconut plantations is the dominant driver for mangrove loss in South Sumatra, unlike most previous studies in Indonesia, in which mangroves were replaced by aquaculture ponds. We also observed that fishpond was one of the drivers for mangrove loss in the study site, but the number was insignificant. The estimated carbon emissions suggests that mangrove conversion to coconut plantation led a higher carbon emissions compared to other land use change, even though this land use was not the largest driver for mangrove loss in ATPF. Nevertheless, our carbon emissions estimate, which employed secondary data and simple back-in-envelope calculations, may generate some uncertainties and limitations, such as unaccounted emissions from other GHGs, as well as aquatic components. These limitations suggest that our findings and associated assumptions can be considered as a preliminary approach, therefore providing a complementary foundation for future studies on mangrove and associated land use emissions in the region. Overall, our findings imply that improved coastal land management with the avoidance of future mangrove loss is urgently required in order to support national efforts for reducing carbon emissions from the land sector.